On Thu, 6 Aug 2020 at 20:14, Naresh Kamboju [off-list ref] wrote:

On Tue, 28 Jul 2020 at 22:04, Christoph Hellwig [off-list ref] wrote:

quoted

Factor out a path_umount helper that takes a struct path * instead of the
actual file name.  This will allow to convert the init and devtmpfs code
to properly mount based on a kernel pointer instead of relying on the
implicit set_fs(KERNEL_DS) during early init.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/namespace.c | 40 ++++++++++++++++++----------------------
 1 file changed, 18 insertions(+), 22 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index 6f8234f74bed90..43834b59eff6c3 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c

<trim>

Regressions on linux next 20200803 tag kernel.
LTP syscalls test umount03 mount a path for testing and
umount failed and retired for 50 times and test exit with warning
and following test cases using that mount path failed.

LTP syscalls tests failed list,
    * umount03
    * umount2_01
    * umount2_02
    * umount2_03
    * utime06
    * copy_file_range01

The reported issue has been fixed in linux next 20200817 tag by
below patch.

fs: fix a struct path leak in path_umount
Make sure we also put the dentry and vfsmnt in the illegal flags and
!may_umount cases.
Fixes: 41525f56e256 ("fs: refactor ksys_umount")

- Naresh